According to the Technical Assistance Research Programs (TARP) studies for the White House Office of Consumer Affairs, only 1-5% of customers actually complain to management level staff about customer service problems. The vast majority of them leave without mentioning anything formally, but often speak about related negative experiences within their personal network. In the highly competitive tech industry, where this customer sentiment spreads far and wide electronically, it is more critical than ever to keep customers happy. This means catching customer dissatisfaction early and often – before complaints arise. Today, the industry’s key metric for success is the Customer Satisfaction (CSAT) score, which relies on surveys that only two per cent of clients respond to. Response bias, poor methodology, goal displacement, and metric gaming add more problems to the limited data available, resulting in a deeply broken system not reflective of the true customer experience. Organizations are in desperate need for a leading indicator that gives them a pulse of the customers’ experience, which in turn can help them deliver positive outcomes and differentiate their products and services from those of their competition.
This was the problem that Sid and Rashmi Bhambhani set out to solve. Sid is a former Director at Symantec and Rashmi worked as a Senior Marketing Manager at Sun Life Financial. They both had years of experience dealing with the problem in large organizations, and had the technical backgrounds to develop an innovative solution using artificial intelligence. They built Summatti, an analytics platform for customer support and service organizations that provides unprecedented real-time insights into the customers’ experience. Summatti follows the customer’s journey across every channel using a powerful artificial intelligence back-end that can analyze voice, tone, sentiment and text. They had one minor challenge: finding enough data.
Existing artificial intelligence methods in this space, particularly in the area of natural language processing (NLP) rely heavily on massive amounts of data to function correctly. Google’s pre-trained model that converts words to vectors was developed on a list of 100 billion words from their News dataset. In Summatti’s case, it isn’t just the amount of text they need to train their model; it is the specific context of the words in question. Words like “crashed,” for example, have entirely different meanings when used in customer support. In order to build out the lexicon of their machine learning algorithms, they needed text focused on this topic specifically – and lots of it.
Sid contacted the Communitech Data Concierge to pull unique information that proved to be invaluable because it had relevant texts, conversations and metrics for what would constitute a ‘good’ experience. Using this data, the lexicon could be developed to a much larger extent, but model developers could also see a labelled metric of “satisfaction.” In early 2019, Summatti was awarded the $50,000 prestigious OCE Customer Demonstration Program grant in partnership with IBM.
Communitech is a partner of Startup HERE Toronto. This article originally appeared on their site.